Document 13548522

advertisement
Joel Emer
November 30, 2005
6.823, L22-1
Vector Computers
Joel Emer
Computer Science and Artificial Intelligence Laboratory
Massachusetts Institute of Technology
Based on the material prepared by
Krste Asanovic and Arvind
Joel Emer
November 30, 2005
6.823, L22-2
Supercomputers
Definition of a supercomputer:
• Fastest machine in world at given task
• A device to turn a compute-bound problem into an I/O
bound problem
• Any machine costing $30M+
• Any machine designed by Seymour Cray
CDC6600 (Cray, 1964) regarded as first supercomputer
Joel Emer
November 30, 2005
6.823, L22-3
Supercomputer Applications
Typical application areas
• Military research (nuclear weapons, cryptography)
• Scientific research
• Weather forecasting
• Oil exploration
• Industrial design (car crash simulation)
• Bioinformatics
• Cryptography
All involve huge computations on large data sets
In 70s-80s, Supercomputer ≡ Vector Machine
Loop Unrolled Code Schedule
loop: ld f1, 0(r1)
ld f2, 8(r1)
ld f3, 16(r1)
ld f4, 24(r1)
add r1, 32
fadd f5, f0, f1
fadd f6, f0, f2
fadd f7, f0, f3
fadd f8, f0, f4
sd f5, 0(r2)
sd f6, 8(r2)
sd f7, 16(r2)
sd f8, 24(r2)
add r2, 32
bne r1, r3, loop
Int1
Int 2
loop:
M1
add r1
ld f1
ld f2
ld f3
ld f4
add r2
sd f5
sd f6
sd f7
sd f8
Schedule
bne
M2
FP+
fadd f5
fadd f6
fadd f7
fadd f8
Joel Emer
November 30, 2005
6.823, L22-4
FPx
Joel Emer
November 30, 2005
6.823, L22-5
Vector Supercomputers
Epitomized by Cray-1, 1976:
• Scalar Unit
– Load/Store Architecture
• Vector Extension
– Vector Registers
– Vector Instructions • Implementation
–
–
–
–
–
Hardwired Control
Highly Pipelined Functional Units
Interleaved Memory System
No Data Caches No Virtual Memory Cray-1 (1976)
Core unit of the Cray 1 computer
Image removed due to copyright restrictions.
To view image, visit http://www.craycyber.org/memory/scray.php.
Joel Emer
November 30, 2005
6.823, L22-6
Joel Emer
November 30, 2005
6.823, L22-7
Cray-1 (1976)
64 Element
Vector Registers
Single Port
Memory
16 banks of
64-bit words
+
8-bit SECDED
( (Ah) + j k m )
(A0)
64
T Regs
Si
Tjk
V0
V1
V2
V3
V4
V5
V6
V7
S0
S1
S2
S3
S4
S5
S6
S7
Vi
V. Mask
Vj
V. Length
Vk
FP Add
Sj
FP Mul
Sk
FP Recip
Si
Int Add
Int Logic
Int Shift
80MW/sec data
load/store
( (Ah) + j k m )
(A0)
320MW/sec
instruction
buffer refill
64
B Regs
Ai
Bjk
NIP
64-bitx16
4 Instruction Buffers
memory bank cycle 50 ns
A0
A1
A2
A3
A4
A5
A6
A7
Pop Cnt
Aj
Ak
Ai
Addr Add
Addr Mul
CIP
LIP
processor cycle 12.5 ns (80MHz)
Joel Emer
November 30, 2005
6.823, L22-8
Vector Programming Model
Scalar Registers
r15
v15
r0
v0
Vector Registers
[0]
[1]
[2]
[VLRMAX-1]
Vector Length Register
Vector Arithmetic
Instructions
ADDV v3, v1, v2
v1
v2
v3
Vector Load and
Store Instructions
LV v1, r1, r2
Base, r1
VLR
Stride, r2
+
+
[0]
[1]
v1
+
+
+
+
[VLR-1]
Vector Register
Memory
Joel Emer
November 30, 2005
6.823, L22-9
Vector Code Example
# Scalar Code
# Vector Code
# C code
LI R4, 64
LI VLR, 64
for (i=0; i<64; i++)
LV V1, R1
C[i] = A[i] + B[i]; loop:
L.D F0, 0(R1)
LV V2, R2
L.D F2, 0(R2)
ADDV.D V3, V1, V2
ADD.D F4, F2, F0
SV V3, R3
S.D F4, 0(R3)
DADDIU R1, 8
DADDIU R2, 8
DADDIU R3, 8
DSUBIU R4, 1
BNEZ R4, loop
Joel Emer
November 30, 2005
6.823, L22-10
Vector Instruction Set Advantages
• Compact
– one short instruction encodes N operations
• Expressive, tells hardware that these N operations:
– are independent
– use the same functional unit
– access disjoint registers
– access registers in same pattern as previous instructions
– access a contiguous block of memory
(unit-stride load/store)
– access memory in a known pattern (strided load/store) • Scalable
– can run same code on more parallel pipelines (lanes)
Joel Emer
November 30, 2005
6.823, L22-11
Vector Arithmetic Execution
• Use deep pipeline (=> fast
clock) to execute element
operations
• Simplifies control of deep
pipeline because elements in
vector are independent (=> no
hazards!)
V
1
V
2
V
3
Six stage multiply pipeline
V3 <- v1 * v2
Vector Instruction Execution
Joel Emer
November 30, 2005
6.823, L22-12
ADDV C,A,B
Execution using
four pipelined
functional units
Execution using one
pipelined functional
unit
A[6]
B[6]
A[24] B[24] A[25] B[25] A[26] B[26] A[27] B[27]
A[5]
B[5]
A[20] B[20] A[21] B[21] A[22] B[22] A[23] B[23]
A[4]
B[4]
A[16] B[16] A[17] B[17] A[18] B[18] A[19] B[19]
A[3]
B[3]
A[12] B[12] A[13] B[13] A[14] B[14] A[15] B[15]
C[2]
C[8]
C[9]
C[10]
C[11]
C[1]
C[4]
C[5]
C[6]
C[7]
C[0]
C[0]
C[1]
C[2]
C[3]
Joel Emer
November 30, 2005
6.823, L22-13
Vector Memory System
Cray-1, 16 banks, 4 cycle bank busy time, 12 cycle latency
• Bank busy time: Cycles between accesses to same bank
Base Stride
Vector Registers
Address
Generator
0 1 2 3 4 5 6 7 8 9 A B C D E F
Memory Banks
+
Joel Emer
November 30, 2005
6.823, L22-14
Vector Unit Structure
Functional Unit
Vector
Registers
Elements
0, 4, 8, …
Elements
1, 5, 9, …
Elements
2, 6, 10, …
Lane
Memory Subsystem
Elements
3, 7, 11, …
T0 Vector Microprocessor (1995)
Joel Emer
November 30, 2005
6.823, L22-15
Lane
Vector register
elements striped
over lanes
[24][25] [26] [27][28]
[16][17] [18] [19][20]
[8] [9] [10] [11][12]
[0] [1] [2] [3] [4]
[29]
[30] [31]
[21]
[22] [23]
[13]
[14] [15]
[5]
[6] [7]
For more information, visit _____________________________________________
http://www.icsi.berkeley.edu/real/spert/t0-intro.html
Vector Instruction Parallelism
Joel Emer
November 30, 2005
6.823, L22-16
Can overlap execution of multiple vector instructions
– example machine has 32 elements per vector register and 8 lanes
Load Unit
load
Multiply Unit
Add Unit
mul
add
time
load
mul
add
Instruction
issue
Complete 24 operations/cycle while issuing 1 short instruction/cycle
Joel Emer
November 30, 2005
6.823, L22-17
Vector Chaining
• Vector version of register bypassing
– introduced with Cray-1
LV
V
1
v1
V
2
V
3
V
4
MULV v3,v1,v2
ADDV v5, v3, v4
Chain
Load
Unit
Memory
Chain
Mult.
Add
V
5
Vector Chaining Advantage
Joel Emer
November 30, 2005
6.823, L22-18
• Without chaining, must wait for last element of result to be
written before starting dependent instruction
Load
Mul
Time
Add
• With chaining, can start dependent instruction as soon as first
result appears
Load
Mul
Add
Joel Emer
November 30, 2005
6.823, L22-19
Vector Startup
Two components of vector startup penalty
– functional unit latency (time through pipeline)
– dead time or recovery time (time before another vector
instruction can start down pipeline)
Functional Unit Latency
R
X
X
X
W
R
X
X
X
W
R
X
X
X
W
R
X
X
X
W
R
X
X
X
W
R
X
X
X
W
R
X
X
X
W
R
X
X
X
W
R
X
X
X
W
R
X
X
X
First Vector Instruction
Dead Time
Dead Time
Second Vector Instruction
W
Joel Emer
November 30, 2005
6.823, L22-20
Dead Time and Short Vectors
No dead time
4 cycles dead time
64 cycles active
Cray C90, Two lanes
4 cycle dead time
Maximum efficiency 94% with 128 element vectors
T0, Eight lanes
No dead time
100% efficiency with 8 element
vectors
Joel Emer
November 30, 2005
6.823, L22-21
Vector Memory-Memory versus Vector
Register Machines
• Vector memory-memory instructions hold all vector operands in main memory
• The first vector machines, CDC Star-100 (‘73) and TI ASC
(‘71), were memory-memory machines
• Cray-1 (’76) was first vector register machine
Vector Memory-Memory Code
Example Source Code
for (i=0; i<N; i++)
{
C[i] = A[i] + B[i];
D[i] = A[i] - B[i];
}
ADDV C, A, B
SUBV D, A, B
Vector Register Code
LV V1, A
LV V2, B
ADDV V3, V1, V2
SV V3, C
SUBV V4, V1, V2
SV V4, D
Joel Emer
November 30, 2005
6.823, L22-22
Vector Memory-Memory vs. Vector
Register Machines
• Vector memory-memory architectures (VMMA) require
greater main memory bandwidth, why?
– All operands must be read in and out of memory
• VMMAs make if difficult to overlap execution of multiple
vector operations, why?
– Must check dependencies on memory addresses
• VMMAs incur greater startup latency
– Scalar code was faster on CDC Star-100 for vectors < 100 elements
– For Cray-1, vector/scalar breakeven point was around 2 elements
⇒ Apart from CDC follow-ons (Cyber-205, ETA-10) all major
vector machines since Cray-1 have had vector register
architectures
(we ignore vector memory-memory from now on)
Automatic Code Vectorization
for (i=0; i < N; i++)
C[i] = A[i] + B[i];
Scalar Sequential Code
load
load
load
add
store
load
load
Iter. 2
add
store
Vectorized Code
load
load
Time
Iter. 1
Iter
.1
Joel Emer
November 30, 2005
6.823, L22-23
load
add
add
store
store
Iter
.2
Vector Instruction
Vectorization is a massive compile-time
reordering of operation sequencing
⇒ requires extensive loop dependence analysis
Vector Stripmining
Joel Emer
November 30, 2005
6.823, L22-24
Problem: Vector registers have finite length
Solution: Break loops into pieces that fit in registers, “Stripmining”
ANDI R1, N, 63
# N mod 64
MTC1 VLR, R1
# Do remainder
for (i=0; i<N; i++)
C[i] = A[i]+B[i]; loop:
LV V1, RA
A B
C
DSLL R2, R1, 3
# Multiply by 8
Remainder
+
DADDU RA, RA, R2 # Bump pointer
LV V2, RB
DADDU RB, RB, R2
64 elements ADDV.D V3, V1, V2
+
SV V3, RC
DADDU RC, RC, R2
DSUBU N, N, R1 # Subtract elements
+
LI R1, 64
MTC1 VLR, R1
# Reset full length
BGTZ N, loop
# Any more to do?
Vector Scatter/Gather
Joel Emer
November 30, 2005
6.823, L22-25
Want to vectorize loops with indirect accesses:
for (i=0; i<N; i++)
A[i] = B[i] + C[D[i]]
Indexed load instruction (Gather)
LV vD, rD
# Load indices in D vector
LVI vC, rC, vD # Load indirect from rC base
LV vB, rB
# Load B vector
ADDV.D vA, vB, vC # Do add
SV vA, rA
# Store result
Vector Scatter/Gather
Joel Emer
November 30, 2005
6.823, L22-26
Scatter example:
for (i=0; i<N; i++)
A[B[i]]++;
Is following a correct translation?
LV vB, rB
LVI vA, rA, vB
ADDV vA, vA, 1
SVI vA, rA, vB
#
#
#
#
Load indices in B vector
Gather initial A values
Increment
Scatter incremented values
Vector Conditional Execution
Joel Emer
November 30, 2005
6.823, L22-27
Problem: Want to vectorize loops with conditional code:
for (i=0; i<N; i++)
if (A[i]>0) then
A[i] = B[i];
Solution: Add vector mask (or flag) registers
– vector version of predicate registers, 1 bit per element
…and maskable vector instructions
– vector operation becomes NOP at elements where mask bit is clear
Code example:
CVM
LV vA, rA
SGTVS.D vA, F0
LV vA, rB
SV vA, rA
#
#
#
#
#
Turn on all elements
Load entire A vector
Set bits in mask register where A>0
Load B vector into A under mask
Store A back to memory under mask
Joel Emer
November 30, 2005
6.823, L22-28
Masked Vector Instructions
Simple Implementation
– execute all N operations, turn off result
writeback according to mask
Density-Time Implementation
– scan mask vector and only execute
elements with non-zero masks
M[7]=1 A[7]
B[7]
M[7]=1
M[6]=0 A[6]
B[6]
M[6]=0
M[5]=1 A[5]
B[5]
M[5]=1
M[4]=1 A[4]
B[4]
M[4]=1
M[3]=0 A[3]
B[3]
M[3]=0
C[5]
M[2]=0
C[4]
M[2]=0
C[2]
M[1]=1
C[1]
A[7]
B[7]
M[1]=1
M[0]=0
C[1]
Write data port
M[0]=0
Write Enable
C[0]
Write data port
Compress/Expand Operations
Joel Emer
November 30, 2005
6.823, L22-29
• Compress packs non-masked elements from one vector register
contiguously at start of destination vector register
– population count of mask vector gives packed vector length
• Expand performs inverse operation
M[7]=1
A[7]
A[7]
A[7]
M[7]=1
M[6]=0
A[6]
A[5]
B[6]
M[6]=0
M[5]=1
A[5]
A[4]
A[5]
M[5]=1
M[4]=1
A[4]
A[1]
A[4]
M[4]=1
M[3]=0
A[3]
A[7]
B[3]
M[3]=0
M[2]=0
A[2]
A[5]
B[2]
M[2]=0
M[1]=1
A[1]
A[4]
A[1]
M[1]=1
M[0]=0
A[0]
A[1]
B[0]
M[0]=0
Compress
Expand
Used for density-time conditionals and also for general selection
operations
Vector Reductions
Joel Emer
November 30, 2005
6.823, L22-30
Problem: Loop-carried dependence on reduction variables
sum = 0;
for (i=0; i<N; i++)
sum += A[i]; # Loop-carried dependence on sum
Solution: Re-associate operations if possible, use binary tree to perform reduction
# Rearrange as:
sum[0:VL-1] = 0
#
for(i=0; i<N; i+=VL)
#
sum[0:VL-1] += A[i:i+VL-1]; #
# Now have VL partial sums in one
do {
VL = VL/2;
sum[0:VL-1] += sum[VL:2*VL-1]
} while (VL>1)
Vector of VL partial sums
Stripmine VL-sized chunks
Vector sum
vector register
# Halve vector length
# Halve no. of partials
Joel Emer
November 30, 2005
6.823, L22-31
A Modern Vector Super: NEC SX-6 (2003)
• CMOS Technology
– 500 MHz CPU, fits on single chip
– SDRAM main memory (up to 64GB)
• Scalar unit
– 4-way superscalar with out-of-order and speculative
execution
– 64KB I-cache and 64KB data cache
• Vector unit
– 8 foreground VRegs + 64 background VRegs (256x64bit elements/VReg)
– 1 multiply unit, 1 divide unit, 1 add/shift unit, 1 logical
unit, 1 mask unit
– 8 lanes (8 GFLOPS peak, 16 FLOPS/cycle)
– 1 load & store unit (32x8 byte accesses/cycle)
– 32 GB/s memory bandwidth per processor
• SMP structure – 8 CPUs connected to memory through crossbar
– 256 GB/s shared memory bandwidth (4096 interleaved banks)
Image removed due
to copyright
restrictions.
Image available in
Kitagawa, K., S.
Tagaya, Y. Hagihara,
and Y. Kanoh. "A
hardware overview of
SX-6 and SX-7
supercomputer." NEC
Research &
Development Journal
44, no. 1 (Jan
2003):2-7.
Joel Emer
November 30, 2005
6.823, L22-32
Multimedia Extensions
•
•
•
•
Very short vectors added to existing ISAs for micros
Usually 64-bit registers split into 2x32b or 4x16b or 8x8b
Newer designs have 128-bit registers (Altivec, SSE2)
Limited instruction set:
– no vector length control
– no strided load/store or scatter/gather
– unit-stride loads must be aligned to 64/128-bit boundary
• Limited vector register length:
– requires superscalar dispatch to keep multiply/add/load units busy
– loop unrolling to hide latencies increases register pressure
• Trend towards fuller vector support in microprocessors
Download